def test_tensor_signal_load_indices():
sig = TensorSignal([(2, 6)], object(), None, (4,), None)
assert np.all(sig.tf_indices == np.arange(*sig.slices[0]))
start, stop = sig.tf_slice
assert start == 2
assert stop == 6
sig = TensorSignal(((2, 3), (2, 3), (3, 4), (3, 4)), object(), None, (4,), None)
assert np.all(sig.tf_indices == [2, 2, 3, 3])
assert sig.tf_slice is None
def test_tensor_signal_broadcast():
sig = TensorSignal([0, 1, 2, 3], object(), None, (4, ), None)
base = np.random.randn(4)
sig_broad = sig.broadcast(-1, 2)
assert sig_broad.shape == (4, 2)
assert sig_broad.key == sig.key
assert np.all(
np.reshape(base[sig_broad.indices], sig_broad.shape) == base[:, None])
sig_broad = sig.broadcast(0, 2)
assert sig_broad.shape == (2, 4)
assert sig_broad.key == sig.key
assert np.all(
np.reshape(base[sig_broad.indices], sig_broad.shape) == base[None, :])
def test_tensor_signal_broadcast():
sig = TensorSignal([0, 1, 2, 3], object(), None, (4,), None, None)
base = np.random.randn(4)
sig_broad = sig.broadcast(-1, 2)
assert sig_broad.shape == (4, 2)
assert sig_broad.key == sig.key
assert np.all(
np.reshape(base[sig_broad.indices], sig_broad.shape) ==
base[:, None])
sig_broad = sig.broadcast(0, 2)
assert sig_broad.shape == (2, 4)
assert sig_broad.key == sig.key
assert np.all(
np.reshape(base[sig_broad.indices], sig_broad.shape) ==
base[None, :])
def test_signal_dict_combine():
minibatch_size = 1
signals = SignalDict(None, tf.float32, minibatch_size)
key = object()
assert signals.combine([]) == []
y = signals.combine([TensorSignal([0, 1, 2], key, None, (3, 2), None),
TensorSignal([4, 5, 6], key, None, (3, 2), None)],
False)
assert y.key is key
assert y.tf_indices is None
assert y.shape == (6, 2)
assert np.all(y.indices == [0, 1, 2, 4, 5, 6])
def test_tensor_signal_getitem():
sig = TensorSignal([1, 2, 3, 4], object(), None, (4, 3), None, None)
sig_slice = sig[:2]
assert np.all(sig_slice.indices == (1, 2))
assert sig_slice.key == sig.key
assert sig_slice.shape == (2, 3)
assert sig[...] is sig
sig_adv = sig[[1, 3]]
assert np.all(sig_adv.indices == (2, 4))
assert sig_adv.shape == (2, 3)
def test_tensor_signal_getitem():
sig = TensorSignal([(1, 6)], object(), None, (4, 3), None)
sig_slice = sig[:2]
assert sig_slice.slices == ((1, 3),)
assert sig_slice.key == sig.key
assert sig_slice.shape == (2, 3)
assert sig[...] is sig
sig_adv = sig[[0, 1, 3, 4]]
assert sig_adv.slices == ((1, 3), (4, 6))
assert sig_adv.shape == (4, 3)
def test_tensor_signal_load_indices():
sess = tf.InteractiveSession()
sig = TensorSignal([2, 3, 4, 5], object(), None, (4,), None, tf.constant)
assert np.all(sig.tf_indices.eval() == sig.indices)
start, stop, step = sess.run(sig.tf_slice)
assert start == 2
assert stop == 6
assert step == 1
sig = TensorSignal([2, 4, 6, 8], object(), None, (4,), None, tf.constant)
assert np.all(sig.tf_indices.eval() == sig.indices)
start, stop, step = sess.run(sig.tf_slice)
assert start == 2
assert stop == 9
assert step == 2
sig = TensorSignal([2, 2, 3, 3], object(), None, (4,), None, tf.constant)
assert np.all(sig.tf_indices.eval() == sig.indices)
assert sig.tf_slice is None
sess.close()
def test_tensor_signal_basic():
# check that indices are read-only
sig = TensorSignal([(0, 2)], None, None, None, None)
with pytest.raises(BuildError, match="Slices are read only"):
sig.slices = ((2, 4),)
with pytest.raises(TypeError, match="does not support item assignment"):
sig.slices[0] = (2, 4)
with pytest.raises(TypeError, match="does not support item assignment"):
sig.slices[0][0] = 2
# check ndim
sig = TensorSignal([(0, 2)], None, None, (1, 2), 1)
assert sig.ndim == 2
def test_tensor_signal_basic():
# check that indices are read-only
sig = TensorSignal([0, 1, 2], None, None, None, None, None)
with pytest.raises(BuildError):
sig.indices = [2, 3, 4]
with pytest.raises(ValueError):
sig.indices[0] = 1
# check ndim
sig = TensorSignal([0, 1, 2], None, None, (1, 2), 1, None)
assert sig.ndim == 2
def test_tensor_signal_reshape():
sig = TensorSignal([1, 2, 3, 4], object(), None, (4, 3), None, None)
with pytest.raises(BuildError):
sig.reshape((100,))
sig_reshape = sig.reshape((6, 2))
assert np.all(sig_reshape.indices == sig.indices)
assert sig_reshape.key == sig.key
assert sig_reshape.shape == (6, 2)
sig_reshape = sig.reshape((6, 2, 1))
assert np.all(sig_reshape.indices == sig.indices)
assert sig_reshape.key == sig.key
assert sig_reshape.shape == (6, 2, 1)
sig_reshape = sig.reshape((-1, 2))
assert sig_reshape.shape == (6, 2)
sig_reshape = sig.reshape((-1,))
assert sig_reshape.shape == (12,)
with pytest.raises(BuildError):
sig.reshape((-1, 5))
with pytest.raises(BuildError):
sig.reshape((-1, -1))
with pytest.raises(BuildError):
sig.reshape((4, 4))
def test_signal_dict_gather():
minibatch_size = 1
var_size = 19
signals = SignalDict(None, tf.float32, minibatch_size)
sess = tf.InteractiveSession()
key = object()
val = np.random.randn(var_size, minibatch_size)
signals.bases = {key: tf.constant(val, dtype=tf.float32)}
x = TensorSignal([0, 1, 2, 3], key, tf.float32, (4,), 1)
with pytest.raises(BuildError):
# indices not loaded
signals.gather(x)
# sliced read
x.load_indices()
assert np.allclose(sess.run(signals.gather(x)), val[:4])
# read with reshape
x = TensorSignal([0, 1, 2, 3], key, tf.float32, (2, 2), 1)
x.load_indices()
assert np.allclose(sess.run(signals.gather(x)),
val[:4].reshape((2, 2, minibatch_size)))
# gather read
x = TensorSignal([0, 1, 2, 3], key, tf.float32, (4,), 1)
x.load_indices()
y = signals.gather(x, force_copy=True)
assert "Gather" in y.op.type
x = TensorSignal([0, 0, 3, 3], key, tf.float32, (4,), 1)
x.load_indices()
assert np.allclose(sess.run(signals.gather(x)),
val[[0, 0, 3, 3]])
assert "Gather" in y.op.type
# reading from full array
x = TensorSignal(np.arange(var_size), key, tf.float32, (var_size,), 1)
x.load_indices()
y = signals.gather(x)
assert y is signals.bases[key]
# reading from strided full array
x = TensorSignal(np.arange(0, var_size, 2), key, tf.float32,
(var_size // 2 + 1,), 1)
x.load_indices()
y = signals.gather(x)
assert y.op.type == "StridedSlice"
assert y.op.inputs[0] is signals.bases[key]
# minibatch dimension
x = TensorSignal([0, 1, 2, 3], key, tf.float32, (4,), 1)
x.load_indices()
assert signals.gather(x).get_shape() == (4, 1)
x = TensorSignal([0, 1, 2, 3], key, tf.float32, (4,), None)
x.load_indices()
assert signals.gather(x).get_shape() == (4,)
sess.close()
def test_signal_dict_scatter():
minibatch_size = 1
var_size = 19
signals = SignalDict(None, tf.float32, minibatch_size)
sess = tf.InteractiveSession()
key = object()
val = np.random.randn(var_size, minibatch_size)
signals.bases = {key: tf.assign(tf.Variable(val, dtype=tf.float32),
val)}
x = TensorSignal([0, 1, 2, 3], key, tf.float32, (4,), None)
with pytest.raises(BuildError):
# assigning to trainable variable
signals.scatter(x, None)
x.minibatch_size = 1
with pytest.raises(BuildError):
# indices not loaded
signals.scatter(x, None)
x.load_indices()
with pytest.raises(BuildError):
# wrong dtype
signals.scatter(x, tf.ones((4,), dtype=tf.float64))
# update
signals.scatter(x, tf.ones((4,)))
y = sess.run(signals.bases[key])
assert np.allclose(y[:4], 1)
assert np.allclose(y[4:], val[4:])
# increment, and reshaping val
signals.scatter(x, tf.ones((2, 2)), mode="inc")
y = sess.run(signals.bases[key])
assert np.allclose(y[:4], 2)
assert np.allclose(y[4:], val[4:])
# recognize assignment to full array
x = TensorSignal(np.arange(var_size), key, tf.float32, (var_size,), 1)
x.load_indices()
y = tf.ones((var_size, 1))
signals.scatter(x, y)
assert signals.bases[key].op.type == "Assign"
# recognize assignment to strided full array
x = TensorSignal(np.arange(0, var_size, 2), key, tf.float32,
(var_size // 2 + 1,), True)
x.load_indices()
y = tf.ones((var_size // 2 + 1, 1))
signals.scatter(x, y)
assert signals.bases[key].op.type == "ScatterUpdate"
sess.close()